A total of 36 participants (n = 21 PD patients; n = 15 healthy control subjects) were recruited for the current study. Patients were categorized into two groups: PD with postural instability (PDPI+; n = 10) and without postural instability (PDPI−; n = 11). All recruitment for the patients was based on the diagnostic criteria recommended by the United Kingdom PD Society Brain Bank. Participants provided written informed consent in accordance with the Declaration of Helsinki. All experimental protocols were approved by the University of South Dakota and the University of Iowa Institutional Review Boards. Severity of PD was assessed by the motor part of the Unified Parkinson's Disease Rating Scale (mUPDRS) (39). PDPI+ participants were selected on the basis of the following criteria: (a) their clinical balance score was greater or equal to five (CBS; sum of mUPDRS items (max. 16) # Leg Agility, # Arising from Chair, # Posture, and # Postural Stability); (b) PDPI+ status was clinically verified by a movement disorders specialist; (c) for subjective confirmation of PDPI+, an unassisted balance task (standing on the foam pad) was performed immediately prior to the study trials. All participants performed the NIH Toolbox Dimensional Change Card Sort (DCCS) test to demonstrate cognitive function (40).

Since we intended to examine every-day postural function and due to the potential fall hazard among unmedicated PD patients (7, 8, 12, 13), all PD participants were treated with their usual prescribed levodopa medication and performed the postural task during “ON” medication, without showing dyskinetic features. Clinical demographics were matched across groups and are summarized in Table 1.

Participants stood quietly on the balance pad (15.5″ L × 12.5″ W × 2.5″ H size) with their feet placed equidistantly to the right and left of the center line of the pad (stance width = ~7.75″) and looked straight ahead during the postural control task. This balance pad was made of thermoplastic elastomer foam. They performed the task without any tactile support, yet a study aide posed behind the subjects would step in to prevent an imminent fall. A triaxial accelerometer (Brain Products) was attached to the left thigh, to collect mediolateral (ML; Y-axis) and anteroposterior (AP; Z-axis) signals (41, 42). Accelerometer sensors can be used for the assessment of balance and postural control in disease conditions (43, 44) and were plugged into the AUX port of the EEG amplifier (Brain Products) for simultaneous recording with the EEG signals. We processed acceleration signals in Matlab (MathWorks) to compute time-dependent changes to measure a more subtle reflection of postural control (44). We converted the unit of acceleration signals from “g” to “m/s²″ by multiplication with 9.8. We down-sampled data to 50 Hz and removed the gravitational factor via the Matlab “detrend” function. Post-hoc filtering employed a 4th order Butterworth filter with a band-pass 0.1–3 Hz. Subsequently, we computed the Euclidean mean signal from both X and Y axes, mean acceleration, root mean square values via the Matlab “rms” function and power of the acceleration power spectrum (0.1–3 Hz) by the “FFT” function. The analysis used both mediolateral and anteroposterior plane channels to compute an ellipse representing the mean explored limits of stability and its respective area under the curve (AUC) (45).

A 64-channel customized EEG cap (Easycap) in conjunction with the Brain Vision amplifier (Brain Products) was used to collect signals at hardware-filtered 0.1 Hz high-pass and a 500 Hz sampling rate from cortical and cerebellar regions in PDPI+, PDPI–, and age-matched control healthy subjects during independent undisturbed stance on the balance pad. The “Pz” electrode was used as reference; “Fpz” electrode was used as ground. Signals from the Fp1, Fp2, FT9, FT10, TP9, and TP10 channels were removed before the preprocessing steps due to regular contamination with mimical and masticatory artifacts, resulting in 59 channels. In addition, a custom mid-cerebellar electrode (Cbz) was placed over the posterior fossa that corresponds to medial aspects of lobules VII, VIII, and IX. EEG signals were divided into consecutive 3 s epochs. The signal from the reference electrode was retrieved using the average reference method. Bad channels and bad epochs were classified using the Matlab “FASTER” and “pop_rejchan” algorithms at default parameters (46). Residual traces of eye movements and other artifacts were removed using independent component analysis via the “ADJUST” algorithm which uses artifact-specific spatial and temporal characteristics to reject eye, muscle, and generic discontinuities. Spectral analysis was implemented on the epoched and preprocessed data using the “pwelch” function. We selected a 1-s time window as segment length, and the number of overlapping samples was set to 50% of the window length. We computed relative spectral power using the mean value from 0.1 to 50 Hz to avoid inter-subject variability. Spectral properties of the lateral motor cortical (C3 and C4), mid-frontal leg premotor area (Cz), and mid-cerebellar (Cbz) signals in the theta-band (4–7 Hz) and beta-band (13–30 Hz) were derived and compared between groups.

Signals from the cerebellar EEG electrodes were obtained from among the occipital regions of the 64-channel setup in accordance with previous studies (34–36). We compared the mid-cerebellar (Cbz) and mid-occipital (Oz) signals among all participants. Additional analyses compared the Cbz signals from the nearby muscle activity via EMG recording during resting-state in 5 PD patients. EMG electrodes were placed above the semispinalis capitis muscle. To differentiate between the postural control and the resting state EEG signals, we recorded EEG signals during a resting-state while participants were sitting on the chair with their eyes open. Resting EEG signals were processed similar to the postural control EEG signals. For both resting-state and postural control EEG data, we collected 3–4 min of continuous data which did not significantly differ in time across groups before and after preprocessing. Given the 3-s epoch length, this resulted in 60–80 epochs/trials per participant.

All statistical analyses were performed using the Statistical toolbox of Matlab. Initially, we performed independent t-tests to compare clinical demographics between PD patients and control subjects, PDPI+ and PDPI–. We performed one-way analysis of variance (ANOVA) tests to compare cognitive (DCCS scores), behavioral (acceleration, rms values, ellipse area under the curve, and power spectral values) and EEG (theta and beta power values) outcomes between all three groups and applied multiple comparisons tests using the Tukey–Kramer approach with an alpha level of <0.05. We measured the effect size with Eta² (η²). We applied the Spearman correlation method to compute the relationship between two variables. Resting-state data were analyzed similar to the postural data. In order to compare between the postural task and resting-state activities, we used a 2 × 2 repeated measures ANOVA with a between-subjects variable (groups) and a within-subjects variable (rest vs. postural task).

We demonstrated the signal quality of the mid-cerebellar (Cbz) signal and volume conduction from the mid-occipital signal by computing the signal similarities between Cbz and Oz using the cross-correlation method to export the amplitude and phase values. A reference EMG lead among PD patients (n = 5) on the splenius capitis et cervicis nearby the Cbz lead was compared to the Cbz resting-state activity via correlation and cross-correlation methods. Moreover, to compare task-related activities between Cz, Oz, and Cbz, we first implemented a one-way ANOVA with multiple comparison tests for each group and subsequently performed 2 × 2 repeated measures ANOVA tests with a between-subjects variable (groups) and a within-subjects variable (electrodes) for theta power.

We initially assessed differences in clinical scores between PDPI+, PDPI–, and healthy control participants (see Table 1 for details). The results from the one-way ANOVA assessing differences in cognitive function through the DCCS task revealed a main effect of “group” [F_(2,33) = 5.23, p = 0.011, η² = 0.24; Supplementary Figure 1A]. Pairwise comparisons demonstrated a difference between PDPI+ and controls (p = 0.01) as well as a trend between PDPI+ and PDPI– participants (p = 0.09). No difference was observed between PDPI– and controls (p = 0.63). Further correlation analyses showed significant negative correlations between DCCS scores and clinical balance scores (rho = −0.8, p < 0.001; Supplementary Figure 1B), as well as disease severity assessed using the mUPDRS (rho = −0.78, p < 0.001; Supplementary Figure 1C). These findings underline the considerable interrelationship between cognitive function, postural instability, and disease severity.

A primary purpose of this study was to assess differences in postural stability between the three groups. Therefore, we performed multiple one-way ANOVAs on measures that assess different aspects of posture including the Euclidean mean signal to measure postural outcomes by mean acceleration. These results demonstrated a main effect of “group” [F_(2,33) = 9.66, p < 0.001, η² = 0.37; Figure 1A]. Pairwise comparisons revealed differences between PDPI+ and controls (p < 0.01) and between PDPI+ and PDPI− (p = 0.02), but no difference between PDPI– and controls (p = 0.42). Subsequently, we used the root mean square to measure the magnitude of the acceleration traces. Similarly, we observed a main effect of “group” [F_(2,33) = 9.17, p = 0.001, η² = 0.36; Figure 1B] with pairwise comparisons showing differences between PDPI+ and controls (p < 0.01) and between PDPI+ and PDPI− (p = 0.02), but not between PDPI– and controls (p = 0.49).

From the mediolateral and anteroposterior accelerometer signals we derived a two-dimensional ellipsis, representing the mean limits of stability. The area under the curve (AUC) for these plots demonstrated a main effect of “group” [F_(2,33) = 7.82, p = 0.002, η² = 0.32; Figure 1C], as well as significant differences between PDPI+ and controls (p < 0.01) and between PDPI+ and PDPI− (p = 0.06), but not between PDPI− and controls (p = 0.34). The power of the acceleration power spectrum was computed between 0.1 and 3 Hz and compared across groups. Similar to aforementioned accelerometer results, a main effect of “group” was observed [F_(2,33) = 5.08, p = 0.012, η² = 0.24; Figure 1D] with differences between PDPI+ and controls (p = 0.02) and between PDPI+ and PDPI– (p = 0.03), but not between PDPI− and controls (p = 1).

We then investigated possible associations between clinical balance scores and accelerometer data (Supplementary Figure 2). There was an association between clinical balance scores and mean acceleration (rho = 0.43, p = 0.05) and power of the acceleration power spectrum (rho = 0.46, p = 0.038). An association trend was observed for clinical balance scores with root mean square measures (rho = 0.41, p = 0.063). Overall, behavioral results underlined the expected differences between PDPI+ and PDPI– participants regarding postural instability.

Given our previous studies showing lower mid-frontal and mid-cerebellar theta-band during cognitive and motor tasks (24, 26), we first examined differences between groups among the electrodes Cz and Cbz. Given the established relationship between beta-band activity and movement activity (25, 47), we also examined differences across the three groups for Cz and Cbz in the beta-band. The results from the one-way ANOVA examining the theta-band over Cz (Figure 2A) demonstrated a main effect of “group” [F_(2,33) = 10.25, p < 0.001, η² = 0.38; Figure 2B]. Pairwise comparisons revealed significant differences between PDPI− and PDPI+ participants (p = 0.03) and between PDPI+ and controls (p < 0.01), but again no difference between PDPI− and controls (p = 0.21). For the beta-band, there were no effects of “group” at Cz [F_(2,33) = 1.21, p = 0.310, η² = 0.07; Figure 2C].

Similarly, when examining the theta-band over mid-cerebellar Cbz (Figure 2D), the one-way ANOVA revealed a main effect of “group” [F_(2,33) = 5.86, p = 0.007, η² = 0.26; Figure 2E] with pairwise comparisons showing differences between PDPI+ and controls (p = 0.01) and between PDPI+ and PDPI− participants (p = 0.03), but no difference between PDPI− and controls (p = 0.93). In the beta-band, no effect of any group was seen at the electrode Cbz [F_(2,33) = 1.39, p = 0.264, η² = 0.08; Figure 2F]. Overall, our EEG results demonstrated that PDPI+ participants had lower mid-frontal and mid-cerebellar theta-band power when performing the postural control task. Furthermore, these changes were absent in the generally movement-related beta-band.

In addition to these bands, we also performed a one-way ANOVA to examine alpha-band differences across groups at the mid-frontal Cz and mid-cerebellar Cbz electrodes. At the mid-frontal Cz electrode, no main effect of group was observed [F_(2,33) = 0.09, p = 0.91]. Similarly, no main effect of group was observed at the mid-cerebellar Cbz electrode [F_(2,33) = 0.19, p = 0.83].

Correlation analyses comparing theta-band power at our mid-frontal and mid-cerebellar electrodes with clinical balance scores and mean acceleration from the postural control task provided specific outcomes. Notably, theta-band power over the mid-frontal electrode Cz correlated with mean acceleration (rho = −0.61, p = 0.004; Figure 3A), but not with clinical balance scores (rho = −0.33, p = 0.15; Figure 3B). By contrast, theta-band power over the mid-cerebellar electrode Cbz did not correlate with mean acceleration (rho = −0.38, p = 0.09; Figure 3C), but did correlate with clinical balance scores (rho = −0.45, p = 0.04; Figure 3D). These results serve to reinforce our established group differences and suggest that mid-frontal and mid-cerebellar theta-band activity is related to multiple assessments of balance.

To determine whether signals were specific to electrode locations Cz and Cbz, we assessed differences across the groups for the electrodes surrounding mid-frontal Cz (left motor cortical C3 and right motor cortical C4) and the electrode above Cbz (mid-occipital Oz). Strikingly, no main effects of “group” in the theta-band were observed over C3 [F_(2,33) = 2.41, p = 0.11, η² = 0.13] or C4 [F_(2,33) = 1.9, p = 0.16, η² = 0.1], or Oz [F_(2,33) = 2.74, p = 0.08, η² = 0.14]. Additionally, no effects were observed in the beta-band for electrodes C3 [F_(2,33) = 0.16, p = 0.85, η² = 0.01], C4 [F_(2,33) = 0.47, p = 0.63, η² = 0.03], or Oz [F_(2,33) = 1.12, p = 0.34, η² = 0.06]. Spectral power topographic plots for each group and frequency band can be observed in Supplementary Figure 3. These results clearly demonstrate the specificity of group differences to the mid-frontal and mid-cerebellar locations and demonstrate that the effects were independent from the surrounding electrodes.

To further distinguish mid-cerebellar activation from the surrounding muscles within the theta-band, we assessed differences between the Cbz electrode and surrounding EMG leads by assessing correlation coefficients (mean ± standard deviation = −0.007 ± 0.02) and zero-lag cross correlations (mean ± standard deviation = 0.00004 ± 0.001; Supplementary Figure 4). Similarly, cross-spectrum phase analyses (mean ± standard deviation = −2.3 ± 7) and cross-correlation analyses (mean ± standard deviation = −5,082.6 ± 70,670.8) were performed to assess similarities between Cbz and the nearby Oz electrode (Supplementary Figure 5). Even though in close proximity to the neck muscles and the Oz electrode, Cbz theta-band activity exhibited distinct activation patterns, as evidenced by the low cross-correlation between electrode Cbz and nearby EMG signals. Additionally, we performed a rmANOVA using electrodes Cbz, Oz, and Cz as within-subjects variables and groups PDPI+, PDPI–, and controls as between-subjects variables to assess theta-band power. The main effect was attributed to “electrode” [F_(2,33) = 32.27, p < 0.001, η² = 0.34] and “group” [F_(2,33) = 8.48, p = 0.001, η² = 0.19], but there was no interaction between the variables [F_(2,33) = 1.27, p = 0.292; Supplementary Figure 6]. Overall, our results describe distinct mid-frontal and mid-cerebellar theta-band changes in PDPI+ participants.

A remaining question was whether these mid-frontal and mid-cerebellar changes were specifically related to the postural task or whether they persisted during rest (Supplementary Figure 7). In contrast to aforementioned correlations in the postural task, a one-way ANOVA examining our mid-frontal site in the resting condition showed no main effect of “group” for the theta-band [F_(2,33) = 0.84, p = 0.44, η² = 0.05] or for the beta-band [F_(2,33) = 0.96, p = 0.39, η² = 0.06]. Similar to the mid-frontal site, no main effects of “group” were observed at the mid-cerebellar site at rest for theta [F_(2,33) = 1.27, p = 0.29, η² = 0.07] and beta [F_(2,33) = 0.97, p = 0.39, η² = 0.06].

Additionally, we performed rmANOVAs with “task” (rest vs. balance task) as within-subjects variables and group as between-subjects variables to assess theta-band and beta-band power. In the mid-frontal theta-band, we found a main effect of “task” (F_(1,33) = 12.34, p = 0.001, η² = 0.23], no main effect of “group” [F_(2,33) = 1.67, p = 0.20, η² = 0.05], and a significant interaction [F_(2,33) = 7.10, p = 0.003]. In contrast, the mid-frontal beta-band showed no main effect of “task” [F_(1,33) = 1.83, p = 0.19, η² = 0.04], no main effect of “group” [F_(2,33) = 1.29, p = 0.29, η² = 0.06], and no interaction [F_(2,33) = 0.04, p = 0.96]. In the mid-cerebellar theta-band, we found no main effect of “task” [F_(1,33) = 0.01, p = 0.94, η² = 0.00] or “group” [F_(2,33) = 0.52, p = 0.6, η² = 0.01], but we did observe an interaction [F_(2,33) = 5.23, p = 0.01]. In the mid-cerebellar beta-band, a trending main effect of “task” was observed [F_(1,33) = 3.95, p = 0.055, η² = 0.07], but no main effect of “group” [F_(2,33) = 1.25, p = 0.3, η² = 0.06] or interaction [F_(2,33) = 0.56, p = 0.57]. Overall, these analyses demonstrate that the specific theta band activity was proprietary to the postural control task and could not be observed in a similar manner during the resting state.